How Enterprises Can Practically Leverage Large Language Models Today Through Platform Partnerships

One of my CTO friends, Mr. Zhang, told me about his painful experience of implementing large models in his company: In June this year, Mr. Zhang wanted to use large models to optimize business processes and replace some repetitive work, so he came to A large model solutions company deployed a private, generic large model to them without understanding their business scenarios.

The large model has been deployed, but it does not solve the problems in the enterprise at all and cannot be used by the business side. Mr. Zhang proposed customized development based on business scenarios, and the other party offered a sky-high price. Mr. Zhang feels that he has been “cheated” and is now riding a tiger, hoping that I can give him some advice. I said, you are following a “pseudo big model”. The fundamental reason is that you don’t know enough about big models.

Einstein said: “The world we create is the product of our thinking. If we don’t change the way we think, we can’t change it.” Next, the author will talk to you about some cognitions and thoughts about “really large models”.


Anyone who tells you that “if you get a big model, you can improve your company’s performance” is a liar

At the beginning of this year, ChatGPT and AIGC detonated the AI circle and became a hot topic in the whole society. Many CTO/CIOs seem to have seen the dawn of GenAI’s ability to reduce costs and increase efficiency for enterprises. In less than a year, the large model has developed from “fashionable” to “really useful”. In addition to writing poetry and drawing, AIGC can also become a productivity tool to help employees in text creation, search, daily office work and application development. Really improve efficiency.

Clayton Christensen, author of “Disruptive Innovation”, once said: “The motivation for disruptive technology is often to solve the simplest problems in the industry. Disruption means that things become simpler and more affordable. Starting.” For enterprises, what is more needed is a large model that is easy to use and affordable.

Although AIGC is developing rapidly, there are still some “uncertainties”, such as high training costs, low business relevance, unstable output content, etc., which are very unfriendly to enterprise applications. In other words, it is difficult for general large models to directly solve practical problems. Enterprises that want to use large models must first solve the two problems of accuracy and security. For now, introducing the PaaS layer is an effective solution.

Take DingTalk AI PaaS as an example. DingTalk has launched an intelligent base AI PaaS for ecological partners and enterprises. It connects large model capabilities to the real needs of users in thousands of industries, further lowering the threshold of AI technology and allowing everyone to use it. The model’s capabilities enter the work scenario and output stably. Based on AI PaaS, enterprises can build exclusive intelligent applications quickly and with low threshold.

At the same time, more and more exclusive and self-built models will emerge in medium and large enterprises. With the accumulation of industry knowledge and supervised fine-tuning, the company’s exclusive large model can provide more accurate and business-valued services for specific scenarios.

Jeffrey Moore, the proposer of “Moore’s Law”, pointed out in “Cross-Domain Chasm”: A new technology will go through a complete “technology adoption life cycle” from invention to widespread application. The initial stage should target those 13.5% of “early adopters”.

Awinic Electronics is one such early user. By introducing the DingTalk AIGC solution, they successfully solved the professional customer service dilemma of 5 major categories, dozens of subcategories, thousands of product SKUs, and tens of thousands of parameters. By using AI solutions to process some complex logic and technical applications internally, Awinic Electronics does not need to do a complete prompt word project, but can also efficiently meet the needs of question and answer scenarios and promote the continuous improvement of customer service experience.

Let’s look back at Mr. Zhang’s case, and it’s not difficult to come to the conclusion: Anyone who tells you that “using a large model can improve the company’s performance” is a liar. The application of AIGC needs to create a dedicated large model based on the company’s business scenarios and build intelligent applications based on the capabilities of AI PaaS. There is no such thing as a large model that can “cure all diseases”.


How to use large models? The attempts of these companies are worth learning from

The outbreak of AIGC has brought some variables to the digital transformation of enterprises, and bosses and business parties have increased expectations for AI. At the same time, there are two obvious changes in digital application construction:

First, AI Agent makes “human-machine collaboration” the norm.

According to IDC’s survey: All companies believe that AI Agent is a relatively certain development direction for large models, 50% of companies are conducting pilot projects, and 34% are making plans. AI Agent has relatively mature application scenarios in enterprises, such as providing schedule reminders, travel arrangements, conference room reservations, text assistants, meeting shorthand, knowledge Q&A and other intelligent functions.

An Xiaopeng, vice president of Alibaba Research Institute and vice president of Alibaba Cloud Intelligence Group, believes: “AI redefines organizational form. ‘A generation of collaboration technology, a generation of organizational form.'” An organizational change has quietly occurred.

“Parkinson’s Law” tells us that in large organizations, the levels of personnel are increasing like a pyramid, the personnel are expanding, everyone is busy, but the efficiency of the organization is getting lower and lower. The emergence of digital employees has made “human-machine collaboration” the norm, making hierarchies flatter, greatly improving the overall operational efficiency of the enterprise, and effectively breaking the curse of “Parkinson’s Law”.

For example, DingTalk cooperated with No. 1 Zhipin to implement digital employee applications in the HR field, integrating AIGC technology to automate a series of tasks in the recruitment and talent management processes. Yihaozhipin no longer builds the APP independently, but innovatively decomposes the backend business processes into different plug-ins and fully integrates them into DingTalk’s capability system, making all links consistent with the usage habits of DingTalk users and making DingTalk more accessible. Nail AIGC achieves fine-grained integration.

The second is that business processes are moving towards “senseless intelligence”.

What is “senseless intelligence”? It is atomized AI capabilities that will be integrated into enterprise business processes at a finer granularity. You can’t feel it, but it is everywhere, just like water to fish and air to humans. AIGC’s capabilities penetrate into every aspect of enterprise operations, replacing a large amount of repetitive work and improving efficiency for enterprises.

“Senseless intelligence” will greatly enhance the software experience. There is a famous “principle of least surprise” in the design field, which means that an excellent experience design should meet most people’s expectations and should not surprise or surprise users. With the help of AIGC, the experience of enterprise intelligent applications will become better and better.

To give a specific case, Tieqilishi cooperates with DingTalk’s AI PaaS and Chat AI products to use intelligent technology to handle massive and complex knowledge consultation from external customers and training needs for the use of internal business systems. Let AI’s question-and-answer results become part of the standardized service process, which not only ensures efficiency and accuracy, but also comprehensively relieves the work intensity and pressure of back-end support staff.

It’s not that large models cannot be used, but they need to be integrated with the company’s data and business.


Enterprise competition is about creativity, and it’s not far off!

Another more obvious problem has also arisen. AI is bringing about an era where ideas are productivity and creativity is productivity. AIGC will significantly lower the threshold for enterprise digitalization, and applications will move from cloud-native to AI-native. Large models and AIGC drivers are redefining infrastructure and AI native design ideas, becoming a new paradigm for software development.

IDC’s research shows that enterprises believe that native AI will bring about a series of changes, including changes in technology stacks, tool chains, infrastructure, development processes, security policies, design concepts, and organizational changes. changes etc.

The software development model based on AIGC allows software engineers to focus more on the data, API and business logic levels. Interface interaction, high-quality code generation, testing and deployment will be completed by AI. In the future, the competition will no longer be about coding speed, but rather the understanding of business, grasp of needs, and how to inspire better ideas.

Einstein said: “Imagination is more important than knowledge. Because knowledge is limited, but imagination is infinite. It contains everything, promotes progress, and is the source of human evolution.”

To put it bluntly: In the AIGC era, creativity is left to humans and development is left to AI. The sweeping lady can also be a great developer.

Driven by AIGC, the application of “+AI” is transformed into “AI+”. What is “AI+”? That is to say, all applications will use AI capabilities as the core driving force, and AI will define scenarios, so that AI practices will be implemented throughout the entire life cycle of business applications.

For example, what bosses and senior management care most about are not cold statistical reports. They want a super assistant who understands business, data, and business operations. Just tell him which indicators to analyze, and he will tell you the results quickly and well, helping you gain deep insights and make decisions that are more beneficial to the development of the enterprise. AI has a very rich application scenario in the field of enterprise decision-making assistance.

Therefore, in the AIGC era, talents with ideas will become increasingly important.


The effectiveness of business management is king, everything else is nonsense

Drucker said: “Efficiency is doing things correctly; effectiveness is doing the right things.” How can contemporary enterprises better solve the effectiveness of organizational collaboration?

I found the answer on DingTalk. As a national-level AI office application, DingTalk is reconstructing itself using large models. At the spring DingTalk Summit held in April this year, DingTalk President Ye Jun said: “We need to redo DingTalk with a large model.” Subsequently, many scenes of DingTalk’s product lines were intelligently recreated in a short period of time. In November, DingTalk officially launched the intelligent office solution “AI Magic Wand”, and 17 product lines of DingTalk have completed intelligent reinvention.

The author is one of the first technology bloggers in China to pay attention to AIGC. I am also very fortunate to be the first batch of internal beta users of DingTalk’s “AI Magic Wand”. DingTalk’s intelligent office capabilities help me manage easily and comprehensively improve the effectiveness of management. . Among the many intelligent products, there are three that impressed me the most:

One is, flash memory. This is the meeting minutes function of DingTalk meetings. You can receive a push flash message after the meeting and automatically generate a meeting minutes. The generated minutes can clearly mark the key points and to-do items of the entire meeting. Even if I don’t participate in the meeting, I can still know the content of the meeting in two or three minutes, which saves the “synchronization” time of the meeting content.

The second is the intelligent question and answer robot. It solves a problem faced by enterprises – activating digital assets. Every enterprise and company will have huge rules and regulations, SOP materials, product materials, etc., but no matter how good the knowledge base is, it will still face an embarrassing problem, that is, “can’t find it”, so the knowledge base cannot find it. It really became a warehouse. By feeding documents, you can train an enterprise-specific Q&A robot. This function can be used in many scenarios, including customer support and employee training. Every authorized member can ask it a question and get a reply.

The third is intelligent low-code construction. This small function solves the problem of casual digitization. Yida AI can generate a software application without writing a line of code. You only need to tell Yida AI your needs through text or voice, or draw the form interface you want, and let Yida use it by taking a photo. Use AI to automatically identify. It can generate the application you want, support modification, and then publish it with one click. After digitization, a large amount of information is also online. Only by training AI can a positive cycle be formed.

These capabilities of DingTalk have helped me a lot in my daily management of the company. I often tell my friends that we are a company driven by DingTalk, and intelligent office applications have greatly improved organizational collaboration. effectiveness.


Kevin Kelly said, “We shape our tools, and our tools shape us.”

As an efficiency tool, AIGC has been integrated into people’s production and life, and is reshaping our world at an unprecedented speed. What will the future hold? There are different opinions. Instead of falling into fantasy and speculation, it is better to be down-to-earth and take action. Because the best way to predict the future is to create it.

error: Content is protected !!